fix: extract test files from git diff format in workspace export

- Fixed _extract_test_files to parse git diff format properly
- Extracts lines starting with '+' (added content) from diffs
- Handles '+++ b/path' to get file paths
- Supports '# Test file:' format from TestGenerator
- Tests are now correctly extracted to tests/ directory

Author: Atlas

src/swe_forge/export/workspace.py

@@ -182,11 +182,71 @@ def export_tasks_to_workspace(
 
 
 def _extract_test_files(test_patch: str, tests_dir: Path) -> None:
-    """Extract test files from diff format."""
-    pattern = r"#\s*(.+\.py)\n(.+?)(?=#\s*.+\.py\n|$)"
-    matches = re.findall(pattern, test_patch, re.DOTALL)
+    """Extract test files from test_patch.
 
-    for file_path, content in matches:
+    Handles multiple formats:
+    1. Git diff format (diff --git ... +lines)
+    2. "# Test file: path" format (from TestGenerator)
+    3. Legacy "# file.py" format
+    """
+    tests_dir.mkdir(parents=True, exist_ok=True)
+
+    # Format 1: Git diff format
+    if "diff --git" in test_patch:
+        diffs = test_patch.split("diff --git")
+
+        for diff in diffs[1:]:
+            plus_match = re.search(r"\+\+\+ b/(.+)", diff)
+            if not plus_match:
+                continue
+            file_path = plus_match.group(1).strip()
+
+            content_lines = []
+            in_hunk = False
+
+            for line in diff.split("\n"):
+                if line.startswith("---") or line.startswith("+++"):
+                    continue
+                if line.startswith("index") or line.startswith("new file"):
+                    continue
+                if line.startswith("@@"):
+                    in_hunk = True
+                    continue
+                if not in_hunk:
+                    continue
+                if line.startswith("+"):
+                    content_lines.append(line[1:])
+
+            if content_lines:
+                while content_lines and not content_lines[-1].strip():
+                    content_lines.pop()
+
+                test_file = tests_dir / Path(file_path).name
+                with open(test_file, "w", encoding="utf-8") as f:
+                    f.write("\n".join(content_lines))
+                    if content_lines:
+                        f.write("\n")
+        return
+
+    # Format 2: "# Test file: path"
+    pattern2 = r"#\s*Test file:\s*(.+)\n(.+?)(?=#\s*Test file:|$)"
+    matches2 = re.findall(pattern2, test_patch, re.DOTALL)
+
+    if matches2:
+        for file_path, content in matches2:
+            file_path = file_path.strip()
+            if not file_path:
+                continue
+            test_file = tests_dir / Path(file_path).name
+            with open(test_file, "w", encoding="utf-8") as f:
+                f.write(content.strip() + "\n")
+        return
+
+    # Format 3: Legacy "# file.py" format
+    pattern3 = r"#\s*(.+\.py)\n(.+?)(?=#\s*.+\.py\n|$)"
+    matches3 = re.findall(pattern3, test_patch, re.DOTALL)
+
+    for file_path, content in matches3:
         file_path = file_path.strip()
         if not file_path:
             continue

test-dataset-v2/TARS-AI-Community-TARS-AI-353/patch.diff

@@ -0,0 +1,63 @@
+diff --git a/src/app.py b/src/app.py
+index cdec0c3d..58418d41 100644
+--- a/src/app.py
++++ b/src/app.py
+@@ -16,6 +16,7 @@
+ import os
+ import sys
+ import threading
++import urllib.request
+ from datetime import datetime
+ 
+ # === Custom Modules ===
+@@ -39,6 +40,33 @@
+ CONFIG = load_config()
+ 
+ # === Helper Functions ===
++def ensure_tars_onnx():
++    """
++    Checks that src/tts/TARS.onnx exists. If not, downloads it from HuggingFace.
++    """
++    onnx_path = os.path.join(BASE_DIR, "src", "tts", "TARS.onnx")
++    onnx_url = "https://huggingface.co/olivierdion007/TARS-AI/resolve/main/TARS.onnx"
++
++    if os.path.isfile(onnx_path):
++        queue_message(f"LOAD: TARS.onnx found at {onnx_path}")
++        return
++
++    queue_message(f"LOAD: TARS.onnx not found. Downloading from HuggingFace...")
++    os.makedirs(os.path.dirname(onnx_path), exist_ok=True)
++
++    try:
++        def _progress(block_num, block_size, total_size):
++            if total_size > 0:
++                pct = min(block_num * block_size / total_size * 100, 100)
++                print(f"\r  Downloading TARS.onnx: {pct:.1f}%", end="", flush=True)
++
++        urllib.request.urlretrieve(onnx_url, onnx_path, reporthook=_progress)
++        print()  # newline after progress
++        queue_message(f"LOAD: TARS.onnx downloaded successfully to {onnx_path}")
++    except Exception as e:
++        queue_message(f"ERROR: Failed to download TARS.onnx: {e}")
++        raise
++
+ def init_app():
+     """
+     Performs initial setup for the application
+@@ -52,6 +80,9 @@ def init_app():
+     if CONFIG['TTS']['ttsoption'] == 'xttsv2':
+         update_tts_settings(CONFIG['TTS']['ttsurl'])
+ 
++    # Ensure TARS voice model is present
++    ensure_tars_onnx()
++
+ def start_discord_in_thread():
+     """
+     Start the Discord bot in a separate thread to prevent blocking.
+@@ -117,4 +148,4 @@ def start_discord_in_thread():
+     finally:
+         stt_manager.stop()
+         bt_controller_thread.join()
+-        queue_message(f"INFO: All threads and executor stopped gracefully.")
+\ No newline at end of file
++        queue_message(f"INFO: All threads and executor stopped gracefully.")

test-dataset-v2/TARS-AI-Community-TARS-AI-353/workspace.yaml

@@ -0,0 +1,16 @@
+task_id: TARS-AI-Community-TARS-AI-353
+repo:
+  url: https://github.com/TARS-AI-Community/TARS-AI.git
+  base_commit: b00108e2a5ac2845d4132532fbcd0c5c4de7adba
+  merge_commit: 0da7e163c6f27e0f32ee59e3d282c9bdde15f7b1
+language: python
+difficulty_score: 1
+prompt: Update app.py
+environment:
+  image: ubuntu:24.04
+  language_version: '3.12'
+install:
+  commands: []
+tests:
+  fail_to_pass: []
+  pass_to_pass: []